Case study
Warehouse analytics, with an assistant that cites its screens
Analytics for a US third-party logistics company, built on its warehouse system, and an AI assistant whose every number links to a screen.
What we built
We built a multi-tenant platform that keeps a mirror of the 3PL’s WMS data and turns it into client scorecards, weekly KPIs with anomaly flags, stock ageing, receivables, a slotting finder and a 3D space heatmap.
On top, an AI assistant answers plain-English questions using only tested read tools, so each answer matches a screen and links to it.
How the assistant answers
-
You ask
A question in plain English, such as “which clients had no picking last month?”
-
It looks it up
The assistant can only call read tools that wrap the functions behind named screens. It can read the data, never change it.
-
It shows its working
Every number in the answer matches a screen, and the answer links to it.
How we test it
The assistant is tested against a dataset with known answers, plus a decoy tenant that reuses the same client codes, so an answer drawn from the wrong tenant fails the test.
Results
- 1,600+automated testsSource: our count of the platform’s test suite, October 2026
Related pages
Get in touch
Want something like this? Tell us the job.
One paragraph is enough. We’ll reply with honest thoughts on whether we can help, and what it would cost.